def process_node(pts, nodeIter):
if len(pts) > 500000:
ptMax = np.amax(pts, axis=0)
ptMin = np.amin(pts, axis=0)
aabbSize = ptMax - ptMin
midPt = (ptMax + ptMin) * 0.5
axis = np.where(aabbSize == np.amax(aabbSize))
axis = axis[0][0]
ptsLeft = np.array([pt for pt in pts if pt[axis] < midPt[axis]])
ptsRight = np.array([pt for pt in pts if pt[axis] >= midPt[axis]])
print("Processing intermediate node")
print(len(pts))
print(len(ptsLeft))
print(len(ptsRight))
print()
maxPtLeft = ptMax
maxPtLeft[axis] = midPt[axis]
minPtRight = ptMin
minPtRight[axis] = midPt[axis]
nodeIter = process_node(ptsLeft, nodeIter)
return process_node(ptsRight, nodeIter)
else:
save_model(args.srcFolder + "/RueMadame_" + str(nodeIter), pts)
print("Saving leaf node")
print(nodeIter)
print()
nodeIter += 1
return nodeIter
inPts: points,
inBatchIds: batchIds,
inFeatures: features,
inCatLabels: catLabels,
inLabels: labels,
isTraining: False,
keepProbConv: 1.0,
keepProbFull: 1.0
})
endTimeMeasure = current_milli_time()
accumTime = accumTime + (endTimeMeasure - startTimeMeasure)
#Save models
if args.saveModels:
save_model(
"savedModels/" + modelsPath[0].replace("/", "-") +
"_sampling_" + str(samp) + "_gt", points, labels, colors,
6)
save_model(
"savedModels/" + modelsPath[0].replace("/", "-") +
"_sampling_" + str(samp) + "_pred", points,
predictedLabelsRes.reshape((-1, 1)), colors, 6)
#Compute IoU
numParts = len(segClasses[cat[catLabels[0][0]][0]])
accumIoU = 0.0
for j in range(numParts):
intersection = 0.0
union = 0.0
currLabel = segClasses[cat[catLabels[0][0]][0]][j]
for k in range(len(labels)):
if labels[k] == predictedLabelsRes[k] and labels[
testDataSet.start_iteration()
for it in range(numTestModels):
points, batchIds, features, gi = testDataSet.get_next_batch()
lossRes, predValsRes = sess.run(
[loss, predVals], {
inPts: points,
inBatchIds: batchIds,
inFeatures: features,
inPts2: points,
inBatchIds2: batchIds,
inFeatures2: features,
inGI: gi,
isTraining: False
})
accumTestLoss += lossRes
save_model(args.outFolder + "/" + str(it) + "_pred", points,
np.clip(predValsRes, 0.0, 1.0))
save_model(args.outFolder + "/" + str(it) + "_gt", points,
np.clip(gi, 0.0, 1.0))
if it % 100 == 0:
visualize_progress(it, numTestModels * args.nExec)
accumTestLoss = accumTestLoss / float(numTestModels * args.nExec)
print("Loss: %.6f" % (accumTestLoss))
distancesRes = \
sess.run(initDistancesGraph,
{inPts:initPoints,
inBatchIds:batchIds,
inFeatures:features,
inVertexs: voxelization[0],
inFaces: voxelization[1],
inFaceIndexs: voxelization[2],
inVoxelIndexs: voxelization[3],
inAABBMin: aabbMinVal,
inCellSizes: [voxelization[5]],
isTraining: False})
distColors = float_to_color_scale(distancesRes, 0.02)
save_model(
args.modelsFolder + "/" + modelName + "_" + modelInstance +
"_c", initPoints, distColors)
elif args.noCompError:
distColors = [[255, 255, 255] for pt in initPoints]
save_model(
args.modelsFolder + "/" + modelName + "_" + modelInstance +
"_c", initPoints, distColors)
accumErrors = []
accumErrorsDist = []
accumErrorsCluster = []
lastDistances = None
for execIter in range(args.numExecs):
minError = 10.0
minErrorDist = 0.0
minErrorCluster = 0.0
mTestDataSet.start_iteration()
while mTestDataSet.has_more_batches():
_, points, batchIds, features, labels, _, sceneName = mTestDataSet.get_next_batch()
currAccWeights = mTestDataSet.get_accuracy_masks(labels)
for iterExec in range(args.nExec):
lossRes, predictedLabelsRes = sess.run([loss, predictedLabels],
{inPts: points, inBatchIds: batchIds, inFeatures: features, inWeights: currAccWeights,
inAccWeights: currAccWeights, inLabels: labels, isTraining: False, keepProbConv: 1.0, keepProbFull: 1.0})
#Save models
if args.saveModels:
save_model("savedModels/"+sceneName[0]+"_gt", points, labels, objColors)
save_model("savedModels/"+sceneName[0]+"_pred", points, predictedLabelsRes.reshape((-1, 1)),
objColors)
labels = labels.reshape((-1))
#Compute IoU
for k in range(len(predictedLabelsRes)):
if labels[k] != 0:
if labels[k] == predictedLabelsRes[k]:
accumIntersection[predictedLabelsRes[k]] = accumIntersection[predictedLabelsRes[k]] + 1.0
accumUnion[predictedLabelsRes[k]] = accumUnion[predictedLabelsRes[k]] + 1.0
else:
accumUnion[labels[k]] = accumUnion[labels[k]] + 1.0
accumUnion[predictedLabelsRes[k]] = accumUnion[predictedLabelsRes[k]] + 1.0
accumGt[labels[k]] = accumGt[labels[k]] + 1.0